What federal buyers need to succeed with AI-enabled procurement

For five years, the federal government has been under a legal mandate to train acquisition professionals in using and buying artificial intelligence (AI). During that time, federal spending on AI has risen precipitously. As AI technology accelerates into the future, buying relevant, rapidly evolving capabilities requires foresight and understanding. Eighty percent of chief procurement officers (CPOs) surveyed plan to deploy AI tools for spend analytics, contract management and supplier selection over the next three years. Leaders are expecting that procurement operations will be radically better and faster than they are today.

Anticipating this, the 2022 Artificial Intelligence Training for the Acquisition Workforce Act requires the Office of Management and Budget (OMB) to provide AI training for procurement professionals. The legislation mandates regular updates addressing capabilities, risks and ethical implications. The proposed 2025 AI and Critical Technology Workforce Framework Act would expand on this.

There is little evidence as yet that any significant portion of the acquisition workforce receives this kind of training.

Meanwhile, the OneGov program is negotiating deals for enterprise chatbot licenses at a dollar or less and relying on employees to experiment with them. As agencies rush to check AI adoption boxes, they're deploying sophisticated technology to an unprepared workforce, while as many as 80 percent of AI initiatives reportedly fail. The solution isn't more chatbots or cheaper licenses — it's effectively training government buyers to use these tools efficiently to transform their work.

Research from RAND and McKinsey documents failure rates from 70 percent to 85 percent for AI initiatives. Organizations that invest in workforce development have been shown to capture more value than those focusing on technology deployment alone. AI implementations fail not from technological limitations but from inadequate workforce preparation.

Current state of federal AI training

The current administration has continued calling for federal workforce AI training. A March 2025 memo from Defense Secretary Hegseth called for acquisition workforce training in buying software — which includes AI — and using commercial buying methods. The General Services Administration (GSA) has established enterprise agreements through its Multiple Award Schedule and OneGov initiative, offering AI from the biggest providers for as little as $1 per agency annually. GSA also is home to the AI Community of Practice, now comprising 12,000+ members across 100+ government organizations.

GSA offers AI training, with acquisition, leadership and policy and technical tracks. As of September 2024, some 12,000 employees had registered for the training, a fraction of the approximately 200,000 acquisition professionals across the Department of Defense (~162,000) and civilian agencies (~48,000). Much of that workforce still needs to better understand what to buy, how to use it and when to transition to new generations of fast-changing emerging technology. People work must precede technology work.

Despite mandates to enhance AI procurement and use, agencies report significant gaps between AI tool availability and workforce readiness.

Patterns in successful federal implementations

Nonetheless, some federal agencies have demonstrated AI success in procurement operations, revealing implementation patterns worth examining.

In 2020, the IRS completed 1,466 contract modifications in 72 hours using AI — work that previously required 1.5 work-years. Former Chief Procurement Officer Shanna Webbers attributed success to months of process analysis and data preparation before technology deployment.

The Army's DORA bot reduced contractor responsibility determinations from two to three hours to five minutes per transaction. The system processes contract actions by automating vendor financial, capability, performance, integrity, accounting and equipment checks.

The Department of Labor's procurement assistant has handled more than 2,300 user interactions, answering routine questions using agency-specific training data. DOL succeeded by limiting scope to well-defined query types rather than attempting comprehensive automation.

These implementations are good examples of “narrow“ AI methods that share decades of experiential learning about the need for:

• Iterative, agile sprints with extended development timelines (typically 12+ months)
• Narrow initial scope, extensive process documentation before automation
• Sustained leadership support.

Generative AI is rapidly advancing to agentic AI. Agentic AI deploys autonomous systems able to make decisions, plan and act with minimal human supervision. It is proactive, adapts in real time and solves complex, multi-step problems by learning from the environment and coordinating multiple AI agents. It offers much greater efficiency but also greater risk management needs, so it is rarely the best way to start experimenting with AI solutions.

Why AI projects fail

Failed implementations of AI reveal recurring patterns:

1. Unreliable or insufficient data derails many initiatives. Procurement data often lacks the consistency, completeness and structure required to train AI. Agencies frequently discover their historical data, collected for compliance rather than analysis, cannot support AI applications. 
2. Unrealistic timelines can bring failure. Leaders expecting immediate returns abandon projects before they mature. Research shows that meaningful AI results require minimum 12-month commitments.
3. Misaligned expectations between technical capabilities of emerging technology and business requirements can create expensive disappointments. Procurement offices expecting AI to handle protest responses or sole-source justifications discover that AI cannot currently replicate the human judgment required to wrestle with complex scenarios.

Hallmarks of effective training 

Successful AI implementations demonstrate that meaningful workforce preparation extends beyond prompt-writing tutorials or vendor demonstrations. Comprehensive staff development builds fundamental capabilities and change-readiness before introducing technology.

Business process analysis and critical thinking skills enable professionals to map workflows, identify inefficiencies and recognize automation opportunities. These competencies prove valuable regardless of the type of technology being adopted.

Data quality and management capabilities help cross-functional teams dissect and evaluate semantic information quality and recognize which processes generate trustworthy data that can prevent costly AI failures.

The ability to evaluate and mitigate risk enables teams to distinguish tasks and projects suitable for AI augmentation from those requiring heavy reliance on human expertise. Market research and initial draft development benefit from AI augmentation acting in the role of a research assistant that learns by doing. Complex negotiations and ethical determinations may well require human experience and judgment.

Technical integration competencies for fast-moving commercial technology streams, along with attractive evolving government procurement methods, ensure acquisition professionals can validate AI outputs while maintaining decision authority. This includes understanding when to accept, modify, or reject AI recommendations based on mission and compliance requirements.

Training principles

Multiple studies agree that certain training approaches consistently produce positive outcomes.

BCG research indicates that leading AI companies invest strategically in fewer, high-priority opportunities and focus 70% of their resources on people and processes, with only 20% on technology and 10% on algorithms. This suggests training should emphasize operations improvement skills before technology introduction — the most leverage lives in people and process. It is not a new finding that the complementary AI technology enables, reinforces and amplifies business improvement initiatives.

Deloitte's AI Academy emphasizes experiential learning through hands-on projects and industry-specific use cases, showing that regular practice and iterative refinement can help apply new skills to real work scenarios. Learning by doing with feedback embeds capabilities more effectively than classroom-style instruction alone. 

McKinsey research shows that successful organizations experiment better, not just more. They start with clear, testable hypotheses and design pilots for learning, not just success. Training should therefore focus on experimenting with and testing a variety of use cases rather than attempting comprehensive transformations.

Federal experience reinforces these findings. Successful offices invest months in preparation, begin with limited pilots, document lessons learned and expand gradually based on measured results. Action training is most effective. Knowledgeable individuals are embedded in acquisition teams to guide and demonstrate how the process should be tailored. This is similar to the Defense Department’s adaptive acquisition framework (AAF), designed to deliver effective, suitable, survivable, sustainable, affordable and timely solutions to military needs. Under the AAF, techniques appropriate to an acquisition problem are introduced with flexibility and agility.

Beyond workforce development

Agencies achieving AI success invest in supporting conditions beyond workforce development, such as:

• Data governance structures that ensure quality inputs multiply training effectiveness. Without reliable data pipelines and validation, even skilled professionals cannot produce valuable AI applications.
• An incremental approach to capturing and sharing lessons learned that accelerates organizational improvement. Documenting successful approaches and failed experiments spreads effective practices and prevents repeated mistakes.

• Extended leadership commitment — especially through problems and difficult periods — that determines ultimate success. The documented 12-month minimum for meaningful results requires sustained support when initial enthusiasm wanes. 
• Cross-functional collaboration that ensures comprehensive implementation. Procurement, IT, legal and program offices must coordinate to address technical, regulatory and operational requirements collectively.

Moving forward

Agencies can continue spraying AI tools across an unprepared staff and praying they learn to use them. But evidence shows that approach will likely end in failure and wasted resources. Alternatively, they should invest now in just-in-time workforce development that addresses process improvement, data literacy and risk assessment before technology introduction.

Organizations that use action learning to educate their workforces in process and workflow analysis, data management and controlled experimentation capture significantly more value than those focusing solely on technology. 

Procurement offices seeking AI transformation should begin with process mapping, not technology selection. They should invest in workforce capabilities, not just software licenses. They should expect extended timelines enabling action learning, not immediate returns. Following these principles, derived from documented successes and failures, offers a reliable pathway to benefiting from AI.

The opportunity for AI-enabled procurement transformation exists. But it begins with preparing people to partner with AI before introducing technology — a sequence that research, federal experience and common sense all validate. 

Timothy Cooke is owner, president and CEO of ASI Government, LLC. ASI Government provides training and support to federal agencies in AI, acquisition and change management. Its ASI Education division delivers the 12-module online AI procurement course, BuyAI, and Pro, a members-only platform offering acquisition training and collaboration.